Every few months we’re told that a new artificial-intelligence model is the smartest ever created.

It scores higher on mathematics.

It writes better computer code.

It performs better on some benchmark most of us have never heard of.

Interesting?

Certainly.

But if you’re running a small business, there’s a much more useful question:

What can it actually do for me?

This week, we received one of the clearest answers yet.

On 3 September 2026, OpenAI launched GPT‑6 Astra, describing it as its most capable model so far.

There are plenty of extraordinary claims surrounding the launch, including a lively debate about whether systems of this level should now be described as artificial general intelligence.

I’m going to leave that argument to somebody else.

Because buried underneath all the talk about AGI is something potentially far more useful for ordinary businesses.

Astra has become considerably better at using a computer.

Not explaining how you should use one.

Actually using one itself. OpenAI

And I think that may ultimately prove more important to many SMEs than another leap in chatbot intelligence.

Until Now, AI Has Mostly Been an Adviser

Think about how most of us have used AI.

You ask:

“How do I create a sales spreadsheet?”

AI explains it.

Or:

“Write a reply to this customer.”

AI writes the reply.

Then you copy it into your email.

Or:

“Research these competitors and tell me what they’re doing.”

AI produces the research.

Then you decide what to do with it.

This is already extremely useful.

But notice something.

You’re still doing the work around the AI.

You open applications.

You copy information.

You paste things.

You update records.

You create files.

You send emails.

You fill in forms.

You transfer the AI’s intelligence into the real world.

That’s the part that’s starting to change.

Astra Doesn’t Just Tell You What to Do

According to OpenAI, GPT‑6 Astra can perform computer tasks including:

• filling in online forms;
• updating customer records in a CRM;
• organising a calendar;
• conducting online research;
• drafting summaries inside email or document applications;
• analysing data;
• generating charts;
• creating websites;
• testing websites;
• installing and testing software;
• troubleshooting problems visible on a computer screen. OpenAI

Read that list again.

Those aren’t traditional chatbot activities.

They’re work activities.

And that distinction is important.

OpenAI’s own testing also reports a substantial improvement in computer-use efficiency over GPT‑5.6 Sol. In its OSWorld 2.0 latency simulations, Astra completed evaluated computer tasks in roughly 40 minutes on average compared with about 75 minutes for GPT‑5.6 Sol, while also achieving a higher score. OpenAI

Benchmarks should always be treated as indicators rather than guarantees of how something will perform in your particular business.

But the direction is clear.

AI is getting better at operating the same digital tools humans use every day.

Imagine This in a Small Business

Let’s forget benchmarks altogether.

Imagine you run a small building company.

You receive an enquiry from somebody interested in an extension.

Today, you might:

open the email;

copy the customer’s details;

enter them into your CRM;

create a follow-up task;

check your calendar;

research the address;

prepare some initial information;

draft a response;

update your spreadsheet.

None of those individual jobs is particularly difficult.

But together they consume time.

Now imagine giving an AI one instruction:

“Process this new enquiry, add the customer to our system, check that we have all the information we need, prepare a draft response and schedule the appropriate follow-up.”

And it does the work across the relevant applications.

That’s a fundamentally different proposition.

The Difference Between Generative AI and Agentic AI

This is a term you’re going to hear increasingly frequently:

Agentic AI.

Traditional generative AI produces something.

Text.

An image.

Music.

Video.

Computer code.

Agentic AI pursues an objective.

You tell it what outcome you want, and it works through the steps required to get there.

GPT‑6 Astra combines improved reasoning with computer use and multi-step workflows, and OpenAI says it can produce finished documents, spreadsheets and presentations while following existing templates and instructions. OpenAI

For businesses, this potentially means moving from:

“Create this for me.”

to:

“Complete this job for me.”

That one-word difference is enormous.

Consider the Office Administrator

Here’s another way of looking at it.

Suppose an employee spends Monday morning doing this:

Download last week’s sales figures.

Put them into a spreadsheet.

Create three charts.

Compare them with the previous month.

Write a summary.

Place everything into the company’s weekly report template.

Email it to the management team.

An advanced AI system increasingly has the ability to handle that as one connected task.

The employee doesn’t disappear.

Instead, their job changes from:

producing the report

to:

reviewing the report and deciding what the figures mean.

That’s exactly where humans should be spending more of their time.

This Could Be Particularly Important for SMEs

Large companies have automated business processes for years.

They’ve employed programmers.

Bought expensive enterprise software.

Built integrations between systems.

Hired consultants.

Small companies generally haven’t had that luxury.

That’s why better computer-using AI could be disproportionately useful to SMEs.

You may not need to commission expensive bespoke automation if an AI can operate the software you already use.

That’s a subtle but potentially enormous change.

The AI doesn’t necessarily require every piece of software to be redesigned specifically for artificial intelligence.

It can increasingly interact with ordinary digital environments more like a person does.

That could make automation accessible to businesses that previously couldn’t justify the cost.

And You May Already Be Paying for Access

OpenAI says Astra is rolling out to ChatGPT Plus, Pro, Business and Enterprise users, as well as through its API, Microsoft Azure and AWS Bedrock. Usage is included within existing subscription allowances, with additional credits available where needed. The rollout began with a limited group of organisations and is expanding rather than appearing simultaneously for everybody. OpenAI

That last point is worth noting.

The launch hasn’t been entirely smooth. Some paying users initially found that Astra wasn’t available despite the broader launch announcement, leading to criticism of the rollout. The Verge

So if you don’t immediately see every Astra capability available in your account, that doesn’t necessarily mean something is wrong.

The rollout is still developing.

Don’t Give AI the Keys to Everything

There is, however, an obvious problem.

The more capable AI becomes at using a computer, the greater the consequences when it gets something wrong.

If an AI writes a poor draft email, you delete it.

If an AI sends the poor email to your biggest customer, that’s more serious.

If it edits the wrong customer record, purchases something you didn’t want, deletes a file or submits incorrect information, the consequences become considerably more important.

This is why the less glamorous part of the Astra announcement deserves attention.

OpenAI says the system includes additional monitoring designed to identify situations where an agent may have misunderstood instructions or gone beyond its intended scope. Supported agent workflows can be paused or stopped so that a human can review what is happening before deciding whether it should continue. OpenAI

That is exactly the type of safeguard businesses should demand from agentic AI.

Because autonomy without oversight isn’t productivity.

It’s risk.

The Best First Tasks Are Boring

Whenever impressive new AI arrives, there’s a temptation to find an equally impressive use for it.

Don’t.

If you’re experimenting with computer-using AI inside a business, start with something gloriously boring.

For example:

Compile our weekly competitor report.

Or:

Organise these customer enquiries into a spreadsheet.

Or:

Prepare this week’s sales presentation using our existing template.

Or:

Research these ten suppliers and produce a comparison.

Choose something repetitive, time-consuming and easy to verify.

Then measure it.

If the task normally takes a member of staff three hours and the AI reduces that to 30 minutes including checking, you have found something useful.

That’s worth far more than an impressive demonstration.

Think in Terms of Outcomes, Not Prompts

There’s another skill businesses will need to learn.

We’ve spent several years talking about prompt engineering.

How do I ask AI the right question?

Agentic systems change that slightly.

The important skill becomes task definition.

What exactly am I asking the system to achieve?

What is it allowed to do?

What information can it access?

When should it stop?

When must it ask me for approval?

What does a successful result look like?

Those questions sound less like using software.

They sound more like managing an employee.

And that may be one of the biggest workplace changes agentic AI introduces.

RealityBreaks Viewpoint

At RealityBreaks, I think GPT‑6 Astra is important, but perhaps not for the reason attracting the biggest headlines.

Whether Astra qualifies as “AGI” is fascinating.

But it’s also a debate ordinary businesses can safely ignore for the moment.

The development worth watching is much more practical:

AI is learning to cross the gap between knowing and doing.

For the first few years of generative AI, humans remained the bridge.

AI created the answer.

We implemented it.

Increasingly, AI can implement it too.

That’s where the productivity opportunity becomes much larger.

But it’s also where responsible use becomes more important.

The best SME implementation won’t be:

“AI can run everything now.”

It will be:

“AI can safely take this particular piece of repetitive work off our hands.”

Then another.

Then another.

Over time, dozens of small tasks could disappear from the working week.

Not because somebody built one enormous automated system.

But because AI gradually became capable of using the ordinary software the business already had.

That, to me, is the real significance of Astra.

Practical Business Takeaway: Find Your 30-Minute Computer Task

Here’s something useful to do this week.

Watch yourself work for one day.

Every time you perform a repetitive computer task, make a note of it.

Not thinking.

Not making important decisions.

Just doing things on the computer.

Examples might include:

• transferring information between systems;
• filling in forms;
• organising files;
• updating spreadsheets;
• preparing routine reports;
• researching standard information;
• formatting documents;
• updating customer records;
• arranging appointments.

Then identify one task that takes approximately 30 minutes.

Write down exactly what you normally do from beginning to end.

That becomes your AI automation test.

Don’t ask:

“Can AI run my company?”

Ask:

“Can AI reliably give me these 30 minutes back?”

If it can, find the next 30 minutes.

For an SME, that is how this technology becomes valuable.

Thirty minutes at a time.

The Bigger Picture

We often imagine the AI revolution arriving dramatically.

Robots walking into offices.

Entire professions disappearing overnight.

Computers suddenly becoming more intelligent than everybody.

The reality may be considerably quieter.

You arrive at work.

The research is already done.

The spreadsheet has been updated.

The customer records are organised.

The presentation has been prepared.

The routine enquiries have been sorted.

And instead of spending the first two hours of the morning administering your business, you spend them running it.

That is the future GPT‑6 Astra makes a little easier to imagine.

And for millions of small businesses, it may ultimately matter far more than whether an AI can pass another intelligence test.

Soft Call to Action

Artificial intelligence is moving incredibly quickly, but you don’t need to understand every new model or technical benchmark to benefit from it.

What matters is recognising when AI becomes capable of solving a genuine problem in your working day.

At RealityBreaks, that’s what we concentrate on: translating major AI developments into practical opportunities that individuals and small businesses can actually understand and use.

Keep following the RealityBreaks AI Blog as we explore where computer-using AI, agents and automation go next—and, more importantly, where they can genuinely save you time.